Digital soil mapping for fire prediction and management in rangelands

Conventional soil mapping

Conventional soil maps provide information about soil properties that relate to soil genesis, morphology, and classification using polygons for spatial representation (Soil Science Division Staff 2017). The mapping scale is generally dependent upon the specific management goals of the project. Soil data available are usually representative values of named soil types, which are aggregated for spatial representation in polygons. This results in soil map units that can have coarse representation of soil property variability within the polygon. A variety of conventional soil maps have been compiled as the Harmonized World Soil Database (HWSD; Table 1) to provide global coverage (Wieder et al. 2014). These polygon-based maps of global extent generally have coarse spatial resolution and somewhat generalized attribute information due to limitations of mapping such large areas. Many regional soil maps cover individual countries at various resolution and detail. In the US, there are three main soil products including the Digital General Soil Map of the United States (STATSGO2; Soil Survey Staff 2018c), Soil Survey Geographic Database (SSURGO; Soil Survey Staff 2018b), and gridded SSURGO (gSSURGO; Soil Survey Staff 2018a) (Table 1). These provide more detailed information than global datasets, with estimated soil properties and interpretations for land management (Soil Science Division Staff 2017). While these maps are produced by expert soil scientists, there are limitations on the quantification of soil property variability and spatial distribution across the landscape.

A major benefit of soil survey information is the variety of interpretations connected to soil map units. For example, SSURGO provides a rating of potential damage by fire that reflects the susceptibility of nutrient, physical, and biotic soil properties to fire (e.g., Fig. 4). This interpretation is a useful tool for prioritizing resources for both preventative management and post-fire assessment (e.g., Burned Area Emergency Rehabilitation). In addition, soil map units are linked to ecological site classifications, which describe soil- and climate-related variations in potential vegetation and its responses to natural and anthropogenic drivers (Moseley et al. 2010). Ecological sites commonly incorporate fire into “state and transition models,” which describe the causes of ecological state change (e.g., large shifts in plant communities) alongside recommendations to manage state change in desired ways (Bestelmeyer et al. 2010). Recommendations associated with fire management may include fire frequencies needed to sustain or alter a given plant community and seed mixes or erosion control strategies to accelerate post-fire recovery. State and transition models can also be used to scale up ecological site information into disturbance response groups that may facilitate post-fire rehabilitation following large fires (Stringham et al. 2016). For rangeland landscapes that are at risk of transitioning to communities dominated by non-native, invasive plants, state and transition models provide valuable site-specific predictions regarding changes in and management of vegetation in the face of changing fire regimes.

Some studies have used static soil properties from conventional soil maps to model wildfires. Levi (2016) found soil water holding capacity and ecological sites to be useful for explaining fire occurrence in desert grasslands of the southwestern US. Dilts and Sibold (2009) explored the use of soil water holding capacity and infiltration rate to model fire, but found insignificant effects and removed the variable from subsequent models, likely a reflection of the coarse-scaled soil information that they derived from the STATSGO2 database. Coarse resolution of soil inputs have also been identified as a limitation for fire prediction models in northern Wisconsin, where Sturtevant (2007) used STATSGO2 to derive soil water holding capacity, drainage class, and hydric soil ratings. Harden et al. (2001) used a simple metric of soil drainage integrated from soil water holding capacity, infiltration rate, and hydraulic conductivity to characterize the state of Alaska, and then related that to wildfire. They determined that more poorly drained areas had more fire activity than better drained areas. However, they used the STATSGO2 database, and soil map units were about six times larger than the fire polygons, which only allowed for relatively simple statistical analyses. We suggest that using soil information that is better matched to the scale of interest for a particular fire modeling application could be more informative. For example, the above studies may have had better relationships between fire predictions and soil properties if they had used more detailed SSURGO data or a suite of digital soil mapping products.

Digital soil mapping

Digital soil mapping is an approach for predicting soil properties or soil types by incorporating measured soil properties at known point locations with environmental covariate layers having continuous spatial coverage (e.g., Landsat satellite data, digital elevation models; McBratney and Santos 2003, Scull et al. 2003). These soil prediction models can utilize simple regression or complex machine learning and generally provide improved estimates of soil properties at a finer spatial scale than currently available soil map products. A tremendous benefit of digital soil mapping models is the ability to produce some measure of model accuracy or confidence that can be incorporated into subsequent models. Digital soil mapping is generally used to predict static soil properties, but these properties can be incorporated with other models to derive more dynamic soil properties.

Digital soil mapping is a practical solution for refining the spatial variability of soil information for large areas. A variety of digital soil maps are currently available including global (SoilGrids; Hengl et al. 2017) and regional (e.g., POLARIS; Chaney et al. 2016) products (Table 1). SoilGrids is a global product of soil property information available in a raster data format with 250 m resolution (Hengl et al. 2017), with recent advances for the continental US that provide data for 100 m pixels (Ramcharan et al. 2018). There are also local scale examples of digital soil mapping that could be useful for fire modeling. For example, the 2009 Hog Fire burned 73 km² in southeastern Arizona, USA, where approximately half of the burned area occurred on national forest land with no published soil survey available (Fig. 4). However, a digital soil map is available from work by Levi (2017) and provides predicted soil components for most of the burned area. A simple intersection of the digital soil map and the Hog Fire boundary indicated that two soil components accounted for 79% of the burned area. One drawback of digital soil mapping data for fire modeling is that knowing about and then accessing localized data may be difficult as there is not currently a clearinghouse or repository of these data. Some review papers offer one mechanism of identifying existing studies (e.g., Grunwald 2009) and the USDA Natural Resources Conservation Service (NRCS) has compiled an annotated bibliography of digital soil mapping projects with NRCS participation (see NRCS 2018), but such lists are often not comprehensive. Some larger-scale projects like the 100 m SoilGrids project are readily available (Table 1). Another challenge to utilizing previously developed digital soil maps is that project objectives may have produced soil property maps that would not be easily translated to relevant fire ecology questions.

Soil moisture mapping

Perhaps the greatest potential for better incorporating soil information into fire modeling is through soil moisture. Soil moisture conditions are tied closely to live fuel moisture content, which is a critical element of wildfire risk models (Qi et al. 2012). The spatial coverage of in situ soil moisture measurements represents only a very small fraction of the landscapes on which wildfires most commonly occur. Efforts to compile these measured data, such as the North American Soil Moisture Database (Quiring et al. 2016), present more opportunities for utilizing soil moisture measurements in fire research. Advancements in soil moisture modeling also provide much needed information for incorporating into an assortment of applications, including fire modeling. A variety of methods exist to predict soil moisture for large spatial areas including cosmic-ray neutron radiation, indirect Global Positioning System signals, remotely sensed land surface temperature measurements, and remote sensing missions specifically designed to measure soil moisture (Ochsner et al. 2013). In recent years, remotely sensed soil moisture data has been used to model fire activity for large areas using several platforms including the European Space Agency’s Soil Moisture and Ocean Salinity mission (SMOS; Chaparro et al. 2016) and National Aeronautics and Space Administration’s (NASA) Gravity Recovery and Climate Experiment (GRACE; Jensen et al. 2018). NASA’s Soil Moisture Active Passive (SMAP) mission also has tremendous potential for providing valuable soil moisture datasets that could be applied to fire prediction models (Entekhabi et al. 2010). Remote sensing missions may offer the greatest potential to inform pre- and post-fire applications for large areas; however, the coarse spatial resolution remains a limitation for landscape scales (Jensen et al. 2018).

In lieu of directly mapping soil moisture, another approach is to use models for predicting soil moisture conditions. For example, Abatzoglou (2013) used a land surface model to derive soil moisture for predicting area burned across the western US. Coops and Waring (2012) derived soil fertility and available soil water holding capacity for forested areas in a large area of western North America by inverting a forest growth model adjusted with remotely sensed leaf area index. Krawchuk (2011) used modeled soil moisture to explore the influence of global resource gradients on fire distribution, and Waring (2016) used soil water balance to model large wildfires across the western US. These proxy measurements of soil properties can be useful for interpreting factors such as soil moisture, leaf area index, and fuel moisture using remote sensing that can also aid in prediction of pre- and post-fire processes.

Fire danger systems

The most immediate benefit of having detailed soil information prior to fire occurrence is the potential for refining fire prediction models of occurrence and burn severity. Improved soil information with spatially explicit estimates of model confidence can allow utilization of more quantitative fire−soil relationships in fire danger systems. Soil moisture models that require physical soil properties for accurate representation of spatiotemporal soil moisture conditions will also benefit the fire modeling community by providing more robust inputs for dynamic fire risk assessment.

Creative applications of these data can facilitate the development of new prediction tools. For example, derivatives of soil moisture related to the fraction of available soil water can be more useful for predicting fire occurrence than actual soil moisture (Krueger et al. 2015, Waring 2016). Incorporating antecedent conditions of soil moisture (Krawchuk 2011) can be refined with more detailed soil property information resulting from digital soil mapping. Applying these drought index concepts specifically to soil moisture conditions and how that relates to the potential of the soil to hold water requires better constraints on soil properties than currently available with conventional soil maps.

Fire ecology

Interpreting complex relationships between soil, vegetation, climate, and management, and the subsequent feedbacks with fire requires spatially explicit information. As we continue to refine our understanding of fire-prone environments and predict the impacts of changing climate and management, there is an increasing need to quantify all factors involved. Conventional soil maps provide valuable information; however, the scale of soil mapping in many forest and rangeland landscapes limits our ability to derive site-specific relationships necessary for advancing the science of fire ecology. Digital soil mapping techniques present an opportunity to better quantify the relationships between soil and fire, which has largely been unexplored.

Digital soil maps can potentially unlock interdisciplinary scientific questions related to fire ecology. For example, a recent study in Alaska used digital soil mapping to predict soil moisture and interpret fire severity. The authors estimated that 90% of the high-severity fire zone lacked permafrost after fire (Brown et al. 2016). Recent changes in the climate of northern latitudes have heightened concern regarding the melting of permafrost and subsequent effects on carbon dynamics and wildfire susceptibility; soil properties play a major role in these processes. A second example of applying digital soil maps could quantify the restoration trajectories of burned areas or design research studies to further investigate fire effects. For example, Nauman and Duniway (2016) developed a detailed soil map of particle size in the soil profile to identify matching soil-geomorphic sites on the Colorado Plateau, USA. They later combined the soil prediction map with other remote sensing data to evaluate the ecological recovery of disturbed sites following oil and gas extraction (Nauman et al. 2017). This same process could be applied to identify similar soil-geomorphic zones for monitoring and comparing burned and unburned areas.

We know that soils interact with climate, vegetation, and management to control the trajectory of post-fire recovery rates for a given landscape. The first-order effects of fire on soil are related to the changes that happen when soil is heated (Massman and Frank 2010), and the degree of alteration for different soil properties is directly related to soil temperatures reached during a fire (Alcaniz et al. 2018); thus, conditions at the time of fire determine the effect of a fire on the soil. Soil heating is also a major factor controlling the recovery of plants following fire because of the effects on existing vegetation and seed bank (Stephan and Miller 2010; Smith and Abella 2014). Dynamic soil properties like soil moisture and organic matter content interact with static soil properties (e.g., texture, pore space) resulting in varying degrees of heat transfer and soil alteration that vary across spatial and temporal scales (Moody et al. 2013). Detailed knowledge of soil conditions, such as those obtained from digital soil mapping, can enable more quantitative interpretations of soil−fire interactions than are currently available with existing soil information.

Soil erosion models

It is well accepted that fire can adversely affect surface soil properties and alter the spatial patterns of soil resources (Allen and Steers 2011, Sankey et al. 2012a, Sankey et al. 2012b). This is important for predicting erosion and revegetation, but conducting post-fire surveys of soil conditions are expensive and often challenging to complete in a timely manner. It is typical to utilize soil surveys to obtain properties related to erosion and revegetation potential and other characteristics (US National Park Service 2006). Soil maps are thus a critical component for post-fire planning and assessment. For example, soil burn severity assessments are commonly linked with existing models that predict post-fire hydrology and erosion using soil property information (Parsons et al. 2010). Numerous models have been used to predict erosion and debris flow following fire including WEPP (Laflen et al. 1997), GeoWEPP (Renschler 2003), ERMiT (Robichaud et al. 2007), and Ravel RAT (Fu 2004), all of which require soil property inputs (Miller et al. 2016). In most cases, conventional soil maps like STATSGO2 and SSURGO are used to derive these inputs; however, digital soil maps can provide more detailed information with spatially explicit representations of model confidence that can subsequently be incorporated into landscape models.

Digital soil mapping can also provide soil information in areas for which conventional soil maps are currently unavailable (e.g., national forests; Fig. 4). One of the largest areas in the US lacking detailed soil survey information (i.e., SSURGO) is Alaska, which has approximately 801 000 km² of rangeland (Fig. 1). Better soils information means better potential to model soil erosion and watershed effects following fire. Understanding the relationships between burn severity and soil properties is an exceptionally high priority for post-wildfire runoff and erosion research (Moody et al. 2013). There is a great need to have a quantitative data set of important soil information (among other data) to facilitate rapid modeling in response to fire (Miller et al. 2016).